• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

I agree Rroff but Amd will not be standing still on that front and Vega should be a monster at async not sure Polaris will get close to 1080 depends on dice of polaris.
 
Farooq said:
Looks like the big gains nvidea said around async was lies barley any difference from Maxwell first benches out in ashes async over at mobi picker
Click to expand...

The performance of the GP104 products relative to Maxwell suggests the uplift is purely clockspeed (as expected, since it's FINFET Maxwell). I would guess the new SW emulation method simply means that enabled asynch shaders / compute in a game engine stops performance from DROPPING like a stone, like it does on Maxwell in most cases ... not improving it far beyond disabled, as it does on GCN.
 
Rroff said:
They've made a few changes to the dma engines, shared memory access, etc. to minimise the impact or need to resort to context switching.

Yes they are still behind AMD in terms of hardware implementation here - but it isn't a case of nVidia are software emulating async or "don't have async" as some claim (and yes as you say the scheduler frontend is done in software - not trying to claim that isn't heh).
Click to expand...

Your first paragraph says what they have done to minimise the impact or the need to resort to context switching then you go on to say it isnt a case of Nvidia are software emulating Async or dont have Async. Context switching is used in Pre-emption which I believe they have re-jigged a different way to minimise the issues. It does say somewhere that, that is indeed what they have done. I still believe they have no Async Compute on pascal and some of the DX11 to DX12 AOTS benches look like that is indeed the case. If they had hardware Async Nvidia would have shouted it from the rooftops....They will when they put it back into Volta, I am sure.

Anyway that wont matter for now as P10 is not competing with 1080. All will be revealed when Vega launches and DX12 becomes more widespread in games. Nvidia have still launched two great cards but I think we will soon see they are not as good as people have been first led to believe from the "Nvidia Show"...regardless of anything AMD put out.;)
 
titaniumx3 said:
Yes definitely. **** it'll do 4k, ultra settings, gamerworks enabled, physx enabled, hairworks enabled, tesselation maxed @ 144fps locked. No problems.
Click to expand...

Oh how helpful your sarcasm is. According to what NVIDIA told the familiar faced youtubers invited to the texas unveiling, it'll be roughly 25% faster than the Titan X in gaming. More with an OC but you can only gain some much from overclocking. Still not enough for 4K 60fps with high settings. I believe the HBM2 enthusiast big boy card may be the first single card to have the muscle for smooth 4K gaming.
 
RanxZy said:
I believe the HBM2 enthusiast big boy card may be the first single card to have the muscle for smooth 4K gaming.
Click to expand...

And then more demanding games will come out :p chasing 4k performance atm is like trying to find a golden egg. I reckon great 4k performance on PC will only come to GPU's just before consoles get it at 30fps.. makes more sense IMO. I mean stable high fps wise.
 
I don't know what the obsession is with 4k when 2K 100hz gsync Superwide ****** all over it ...

980ti on my X34 is good but its not perfect...

4k 60hz 16.9 sounds lame.
 
Depends on what you are playing and personal taste - personally I can't see anything topping 2560x1440@120+Hz for many multiplayer games, super wide is pretty decent for many single player games the only time I'd use 4K is with stuff like racing games where I'm using a controller.
 
Rroff said:
Depends on what you are playing and personal taste - personally I can't see anything topping 2560x1440@120+Hz for many multiplayer games, super wide is pretty decent for many single player games the only time I'd use 4K is with stuff like racing games where I'm using a controller.
Click to expand...

16:9 now feels antique...

I have 2 Dell U2715H in my studio (just bought them ) and they are great for work...

Gaming 16:9 is naff

The X34 predator is expensive but if you get a good one its an amazing monitor.

I'll probably break and get a 1080 and hopefully it will make the minimums less noticeable...

But from a 980ti it will be all placebo :p
 
Last edited:
You can't say they have no async compute :P you can say they have problems with async compute but its perfectly possible to despatch separate, parallel, 1x graphic + multiple compute queues - how well that actually works is another matter but at the very least if the programmer implements their workload in an optimal manner it is possible to get gains out of it (and on Maxwell it is very easy to break it very badly and even stall the whole OS).
 
Last edited:
Nvidia are claiming big async improvements. This is in their official press release:
New asynchronous compute advances improve efficiency and gaming performance
Click to expand...

A reduction in context switch costs is a significant improvement. We know know from the GP100 nVida improved preemption a lot. If context switch costs cam be reduced enough it will happily outperform any current GCN implementation Arion which has high pverheads. Even with Maxwll this could be seen in the Fabled.legends bechamrks where the 980ti handily beat the Furyx in async tasks. The problem maxwell has is it takes a lot if developer and driver care to avoid expensive context switches.

And for the 1 millionth time, even Maxwell does async compute in hardware. Only the scheduler is driver based.
 
andybird123 said:
to be fair, the 4th position card was $299 10 years ago, $299 to $379 over 10 years is not that bad as far as inflation goes. Its not actually nvidia's fault that the exchange rate sucks at the moment.

the best card 10 years ago was $829, which is well north of $1000 at todays rates and would be around £910 at todays exchange rates and VAT, but people seem to think GPU's have massively increased in price
Click to expand...

I'm still going to moan, I don't care for your your facts and well written explanation :p
 
Found another GTX 1080 2K benchmark with Batman Arkham Knight.


I don't know if it fake or real just like the other video with 4 games, everybody screamed FAKE at GTX 1080 cooler leaked back in March so it turned out real at launch.
 
Wonder what 2017 will bring, shrink (10nm) or refresh (GM200/GK110)?

Pascal pretty underwhelming so far, especially given the shrink + "new" arch. Maybe they need more time to get it right and will bring out a refresh to cover the next year or so. Or just cut their losses and get on to Volta finally which is what they have been working towards the past decade.
 
You must log in or register to reply here.
Back
Top Bottom